In the manufacture of integrated circuits, deposition of thin films of many pure and compound materials is necessary, and many techniques have been developed to accomplish such depositions. In recent years the dominant technique for deposition of thin films in the art has been chemical vapor deposition (CVD), which has proven to have superior ability to provide uniform even coatings, and to coat relatively conformally into vias and over other high-aspect and uneven features in wafer topology. As device density has continued to increase and geometry has continued to become more complicated, even the superior conformal coating of CVD techniques has been challenged, and new and better techniques are needed.
The approach of a variant of CVD, Atomic Layer Deposition has been considered for improvement in uniformity and conformality, especially for low temperature deposition. However the practical implementation of this technology requires a solution to higher purity and higher throughput. This patent addresses these requirements.
Atomic Layer Deposition
In the field of CVD a process known as Atomic Layer Deposition (ALD) has emerged as a promising candidate to extend the abilities of CVD techniques, and is under rapid development by semiconductor equipment manufacturers to further improve characteristics of chemical vapor deposition. ALD is a process originally termed Atomic Layer Epitaxy, for which a competent reference is: Atomic Layer Epitaxy, edited by T. Suntola and M. Simpson, published by Blackie, Glasgo and London in 1990. This publication is incorporated herein by reference.
Generally ALD is a process wherein conventional CVD processes are divided into single-monolayer deposition steps, wherein each separate deposition step theoretically goes to saturation at a single molecular or atomic monolayer thickness, and self-terminates.
The deposition is the outcome of chemical reactions between reactive molecular precursors and the substrate. In similarity to CVD, elements composing the film are delivered as molecular precursors. The net reaction must deposit the pure desired film and eliminate the "extra" atoms that compose the molecular precursors (ligands). In the case of CVD the molecular precursors are fed simultaneously into the CVD reactor. A substrate is kept at temperature that is optimized to promote chemical reaction between the molecular precursors concurrent with efficient desorption of byproducts. Accordingly, the reaction proceeds to deposit the desired pure film.
For ALD applications, the molecular precursors are introduced into the ALD reactor separately. This is practically done by flowing one precursor at a time, i.e. a metal precursor --ML.sub.x (M=Al, W, Ta, Si etc.) that contains a metal element --M which is bonded to atomic or molecular ligands --L to make a volatile molecule. The metal precursor reaction is typically followed by inert gas purging to eliminate this precursor from the chamber prior to the separate introduction of the other precursor.
This purge step (or sometimes a pump-down step) is key for ALD films without the undesired CVD component. During the purge (evacuation) step the last used chemical is removed from the chamber and gas introduction lines, enabling introduction of a different chemical.
An ALD reaction will take place only if the surface is prepared to react directly with the molecular precursor. Accordingly the surface is typically prepared to include hydrogen-containing ligands --AH that are reactive with the metal precursor. Surface--molecule reactions can proceed to react with all the ligands on the surface and deposit a monolayer of the metal with its passivating ligand: substrate-AH+ML.sub.x.fwdarw.substrate--AML.sub.y +HL, where HL is the exchange reaction by-product. During the reaction the initial surface ligands --AH are consumed and the surface becomes covered with L ligands, that cannot further react with the metal precursor --ML.sub.x. Therefore, the reaction self-saturates when all the initial ligands are replaced with --ML.sub.y species.
After completing the metal precursor reaction all of the metal precursor is typically removed from the reactor prior to the introduction of another precursor. The second type of precursor is used to restore the surface reactivity towards the metal precursor, i.e. eliminating the L ligands and redepositing AH ligands.
Most ALD processes have been applied to deposit compound films. In this case the second precursor is composed of a desired (usually nonmetallic) element --A (i.e. O, N, S), and hydrogen using, for example H.sub.2 O, NH.sub.3, or H.sub.2 S. The reaction: --ML+AH.sub.z.fwdarw.--M--AH+HL (for the sake of simplicity the chemical reactions are not balanced) converts the surface back to be AH-covered. The desired additional element --A is deposited and the ligands L are eliminated as volatile by-product. Again, the reaction consumes the reactive sites (this time the L terminated sites) and selfsaturates when the reactive sites are entirely depleted.
The sequence of surface reactions that restores the surface to the initial point is called the ALD deposition cycle. Restoration to the initial surface is the keystone of ALD. It implies that films can be layered down in equal metered sequences that are all identical in chemical kinetics, deposition per cycle, composition and thickness. Self-saturating surface reactions make ALD insensitive to transport nonuniformity either from flow engineering or surface topography (i.e. deposition into high aspect ratio structures). Non uniform flux can only result in different completion time at different areas. However, if each of the reactions is allowed to complete on the entire area, the different completion kinetics bear no penalty.
As is often the case with process development, the initial promised advantages of a new technique do not, in the end, attain their full initial promise. Unfortunately, ALD has a serious fundamental problem. Unlike CVD reactions that are of a continuous steady state nature, ALD reactions follow kinetics of molecular-surface interaction. Kinetics of molecular-surface reactions depends on the individual reaction rate between a molecular precursor and a surface reactive site and the number of available reactive sites. As the reaction proceeds to completion, the surface is converted from being reactive to non-reactive. As a result the reaction rate is slowing down during the deposition. In the simplest case the rate, dN/dt is proportional to the number of reactive sites, dN/dt=-kN, where N is the number of reactive sites and k is the (single site) reaction rate. Eliminating reactive sites (or growing of the already-reacted sites) follows an exponential time dependence kN(t)=kN.sub.0 exp(-kt). This fundamental property of molecule-surface kinetics was named after the great scientist Langmuir, and is quite well-known in the art.
The interpretation of Langmuirian kinetics limitations illustrates a serious drawback of ALD and a severe deviation from the ideal picture. Accordingly, the self-terminating reactions never really self-terminate (they would require an infinite time because the rate is exponentially decreasing). It means that under practical conditions the surface is never entirely reacted to completion after a deposition cycle. If the surface is not completely reacted there are leftover undesired elements in the film. For example, if the ML.sub.x reaction cannot totally consume the surface --AH sites, the film will have H incorporation. Likewise, if the AH.sub.y reaction is not carried to completion, undesired L incorporation is inevitable. Clearly, the quality of the film depends on the impurity levels. The throughput-quality tradeoff is of particular concern because it carries an exponential throughput penalty to attain a reduction of impurity levels.
In conventional atomic layer deposition one must accept low throughput to attain high-purity film, or accept lower-purity films for higher throughput. What is clearly needed is an apparatus and methods which not only overcome the Langmuirian limitations but simultaneously provide higher-purity films than have been available in the prior art methods. Such apparatus and methods are provided in embodiments of the present invention, taught in enabling detail below.
In addition to the above ideal situation, ALD chemicals, such as the ML.sub.x and AH.sub.z in the above example are typically extremely reactive, and will lead to extensive undesired CVD side reactions if they coexist in the chamber even at trace levels. Since CVD is a very undesirable companion, fast and efficient purge has been the most difficult and challenging aspect of engineering high throughput ALD apparatuses.
Chemical delivery lines must be short and free of trapped volume to facilitate efficient purging of chemicals. However, some limitation on efficient purge come from line surface outgassing that is difficult to avoid. Accordingly, some trace of chemical mixing is impossible to suppress with throughput limited short purge times. What is needed is a rapid method of removing trace quantities of the previously used chemical precursor prior to introduction of the desired new chemical precursor.
Our invention, which provides the clear and present need, provides an ALD Pre-Reactor as an apparatus and process that eliminates trace amounts of chemical mixing without CVD contribution to the ALD film on the substrates.